VSORA SA has obtained €40 million (about US$46 million) in funding to enable it to take an AI inference processor into 5nm production with TSMC in 2025.
www.eenewsanalog.com, Apr. 29, 2025 –
The funding was led by Otium and the office of an unnamed wealthy French family. Additional funding was provided by Omnes Capital and Adélie Capital with co-financing from the European Innovation Council (EIC) fund.
VSORA (Paris, France) was founded by CEO Khaled Maalej in 2015 to develop DSP IP cores for chipmakers that were supporting digital communications systems, such as 5G.
The latest round of funding is intended to support the Jotunn 8 (J8) processor, which the company claims will deliver 3x the performance of existing AI inference processors while consuming half the power. The specific AI inference processor being benchmarked against is not disclosed. In a Linkedin posting Maalej said the J8 chip delivers 3,200 teraflops which is the same as the best-in-class chips while consuming less than half the power.
In email VSORA told eeNews Europe that the perfomance claims are based on published MLPerf 4.0 data center inference data when implementing the Llama2-70B LLM.
“This funding marks a pivotal moment for VSORA as we accelerate our mission to revolutionize AI chips and ensure Europe’s technological sovereignty in AI computing,” said Maalej, in a statement.
When operating on tensor cores J8 achieves 800 TFLOPS using an FP16 datatype and 3200 TFLOPS with FP32 datatype. The design is host processor agnostic and includes RISC-V cores to run AI on-chip. The component is designed for 288Gbytes of HBM3e stacked DRAM with a throughput of 8Tbytes per second.
The J8 AI processor has a proprietaty software development platform that uses standard high-level entry points, such as ONNX, Pytorch and others, and then a dedicated LLVM compiler. (ONNX, Pytorch,…) and LLVM compiler. This is the same starting point as the current market leader is using,said Jan Pantzar, vice president of sales and marketing at VSORA, making it straightforward to port existing algorithms to VSORA hardware.
In 2023 VSORA announced the Jotunn 4 chiplet-based although without information on the manufacturing process the design was targeting. It also announced €13 million in funding earmarked for development phase of the project.
In January 2022 VSORA introduced the Tyr series of chips for autonomous driving saying they would sample in 4Q22 and be available in-vehicle in 2024.